الفهرس 
COVEROnly 14 pages are availabe for public view
 |  | from | 32 |  |  |
| | | from | 32 | | |
Abstract
In this thesis, mobile health system architecture for neuromuscular disorders patient emergency was proposed. Also a new hybrid classification technique for neuromuscular disorders diagnosis based on EMG signal processing for the mobile health application was designed.
The proposed technique can be broken down into three important phases, (1) EMG signal acquisition and signal processing phase. (2) Feature extraction and selection phase. (3) Constructing the hybrid classifier.
Through this research, different SVM and ANN classifiers were designed and their performances were verified on different datasets. First dataset is EMG physical action Data set from the machine learning repository (UCI). Ten different normal and aggressive muscles ’physical actions were selected randomly to be classified. The actions were divided into five group each group has two actions, kneeing and pulling, Hammering and Header, Clapping and Handshaking and Elbowing and Slapping. SVM with Different kernel functions and different parameters was applied on each group and a comparison was made between their classification accuracy. The results showed that that SVM with RBF kernel function with sigma equal 1 obtained the best classification accuracy of 90% for kneeling and pulling, 82% for hammering and header, 94% for clapping and handshaking, 98% for running and hugging and 83% for elbowing and slapping.
Also feed-forward NN with LM back-propagation algorithm was used to classify the five groups of actions. The performance of the application of this algorithm with 10 neurons in the hidden layers is compared with the performance of SVM applied on the same dataset using RBF kernel function with sigma equal 1.
The application of LM based neural network classifier achieved accuracy of 90%